Whamcloud - gitweb
b=21919 Do not resend quota_ctl requests
[fs/lustre-release.git] / lustre / quota / quota_ctl.c
index c056048..7c9a66a 100644 (file)
@@ -1,19 +1,42 @@
 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
  * vim:expandtab:shiftwidth=8:tabstop=8:
  *
- *  lustre/quota/quota_ctl.c
+ * GPL HEADER START
  *
- *  Copyright (c) 2005 Cluster File Systems, Inc.
+ * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  *
- *   This file is part of Lustre, http://www.lustre.org.
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 only,
+ * as published by the Free Software Foundation.
  *
- *   No redistribution or use is permitted outside of Cluster File Systems, Inc.
+ * This program is distributed in the hope that it will be useful, but
+ * WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License version 2 for more details (a copy is included
+ * in the LICENSE file that accompanied this code).
  *
+ * You should have received a copy of the GNU General Public License
+ * version 2 along with this program; If not, see
+ * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
+ *
+ * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
+ * CA 95054 USA or visit www.sun.com if you need additional information or
+ * have any questions.
+ *
+ * GPL HEADER END
+ */
+/*
+ * Copyright  2008 Sun Microsystems, Inc. All rights reserved
+ * Use is subject to license terms.
+ */
+/*
+ * This file is part of Lustre, http://www.lustre.org/
+ * Lustre is a trademark of Sun Microsystems, Inc.
  */
 #ifndef EXPORT_SYMTAB
 # define EXPORT_SYMTAB
 #endif
-#define DEBUG_SUBSYSTEM S_MDS
+#define DEBUG_SUBSYSTEM S_LQUOTA
 
 #ifdef __KERNEL__
 # include <linux/version.h>
 # include <linux/init.h>
 # include <linux/fs.h>
 # include <linux/jbd.h>
-# include <linux/ext3_fs.h>
 # include <linux/quota.h>
-# if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
-#  include <linux/smp_lock.h>
-#  include <linux/buffer_head.h>
-#  include <linux/workqueue.h>
-#  include <linux/mount.h>
-# else
-#  include <linux/locks.h>
-# endif
+# include <linux/smp_lock.h>
+# include <linux/buffer_head.h>
+# include <linux/workqueue.h>
+# include <linux/mount.h>
 #else /* __KERNEL__ */
 # include <liblustre.h>
 #endif
 #include <lustre_quota.h>
 #include "quota_internal.h"
 
+#ifdef HAVE_QUOTA_SUPPORT
 #ifdef __KERNEL__
-int mds_quota_ctl(struct obd_export *exp, struct obd_quotactl *oqctl)
+
+int mds_quota_ctl(struct obd_device *obd, struct obd_export *unused,
+                  struct obd_quotactl *oqctl)
 {
-        struct obd_device *obd = exp->exp_obd;
+        struct obd_device_target *obt = &obd->u.obt;
+        struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
+        struct timeval work_start;
+        struct timeval work_end;
+        long timediff;
         int rc = 0;
         ENTRY;
 
+        cfs_gettimeofday(&work_start);
         switch (oqctl->qc_cmd) {
         case Q_QUOTAON:
+                oqctl->qc_id = obt->obt_qfmt; /* override qfmt version */
                 rc = mds_quota_on(obd, oqctl);
                 break;
         case Q_QUOTAOFF:
-                mds_quota_off(obd, oqctl);
+                oqctl->qc_id = obt->obt_qfmt; /* override qfmt version */
+                rc = mds_quota_off(obd, oqctl);
                 break;
         case Q_SETINFO:
                 rc = mds_set_dqinfo(obd, oqctl);
@@ -74,6 +102,13 @@ int mds_quota_ctl(struct obd_export *exp, struct obd_quotactl *oqctl)
         case Q_GETOQUOTA:
                 rc = mds_get_obd_quota(obd, oqctl);
                 break;
+        case LUSTRE_Q_INVALIDATE:
+                rc = mds_quota_invalidate(obd, oqctl);
+                break;
+        case LUSTRE_Q_FINVALIDATE:
+                oqctl->qc_id = obt->obt_qfmt; /* override qfmt version */
+                rc = mds_quota_finvalidate(obd, oqctl);
+                break;
         default:
                 CERROR("%s: unsupported mds_quotactl command: %d\n",
                        obd->obd_name, oqctl->qc_cmd);
@@ -84,67 +119,117 @@ int mds_quota_ctl(struct obd_export *exp, struct obd_quotactl *oqctl)
                 CDEBUG(D_INFO, "mds_quotactl admin quota command %d, id %u, "
                                "type %d, failed: rc = %d\n",
                        oqctl->qc_cmd, oqctl->qc_id, oqctl->qc_type, rc);
+        cfs_gettimeofday(&work_end);
+        timediff = cfs_timeval_sub(&work_end, &work_start, NULL);
+        lprocfs_counter_add(qctxt->lqc_stats, LQUOTA_QUOTA_CTL, timediff);
 
         RETURN(rc);
 }
 
-int filter_quota_ctl(struct obd_export *exp, struct obd_quotactl *oqctl)
+int filter_quota_ctl(struct obd_device *unused, struct obd_export *exp,
+                     struct obd_quotactl *oqctl)
 {
         struct obd_device *obd = exp->exp_obd;
         struct obd_device_target *obt = &obd->u.obt;
         struct lvfs_run_ctxt saved;
+        struct lustre_quota_ctxt *qctxt = &obt->obt_qctxt;
+        struct lustre_qunit_size *lqs;
+        void *handle = NULL;
+        struct timeval work_start;
+        struct timeval work_end;
+        long timediff;
         int rc = 0;
         ENTRY;
 
+        cfs_gettimeofday(&work_start);
         switch (oqctl->qc_cmd) {
         case Q_QUOTAON:
+                oqctl->qc_id = obt->obt_qfmt;
+                rc = generic_quota_on(obd, oqctl, 0);
+                break;
+        case Q_FINVALIDATE:
         case Q_QUOTAOFF:
-                if (!atomic_dec_and_test(&obt->obt_quotachecking)) {
-                        CDEBUG(D_INFO, "other people are doing quotacheck\n");
-                        atomic_inc(&obt->obt_quotachecking);
+                cfs_down(&obt->obt_quotachecking);
+                if (oqctl->qc_cmd == Q_FINVALIDATE &&
+                    (obt->obt_qctxt.lqc_flags & UGQUOTA2LQC(oqctl->qc_type))) {
+                        CWARN("quota[%u] is on yet\n", oqctl->qc_type);
+                        cfs_up(&obt->obt_quotachecking);
                         rc = -EBUSY;
                         break;
                 }
+                oqctl->qc_id = obt->obt_qfmt; /* override qfmt version */
         case Q_GETOINFO:
         case Q_GETOQUOTA:
         case Q_GETQUOTA:
                 /* In recovery scenario, this pending dqacq/dqrel might have
                  * been processed by master successfully before it's dquot
-                 * on master enter recovery mode. We must wait for this 
+                 * on master enter recovery mode. We must wait for this
                  * dqacq/dqrel done then return the correct limits to master */
                 if (oqctl->qc_stat == QUOTA_RECOVERING)
-                        qctxt_wait_pending_dqacq(&obd->u.obt.obt_qctxt,
-                                                 oqctl->qc_id, oqctl->qc_type, 
-                                                 1);
+                        handle = quota_barrier(&obd->u.obt.obt_qctxt, oqctl, 1);
 
                 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
-                rc = fsfilt_quotactl(obd, obd->u.obt.obt_sb, oqctl);
+                rc = fsfilt_quotactl(obd, obt->obt_sb, oqctl);
                 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 
-                if (oqctl->qc_cmd == Q_QUOTAON || oqctl->qc_cmd == Q_QUOTAOFF) {
-                        if (!rc)
-                                obt->obt_qctxt.lqc_status = 
-                                        (oqctl->qc_cmd == Q_QUOTAON) ? 1 : 0;
-                        atomic_inc(&obt->obt_quotachecking);
+                if (oqctl->qc_stat == QUOTA_RECOVERING)
+                        quota_unbarrier(handle);
+
+                if (oqctl->qc_cmd == Q_QUOTAOFF ||
+                    oqctl->qc_cmd == Q_FINVALIDATE) {
+                        if (oqctl->qc_cmd == Q_QUOTAOFF) {
+                                if (!rc)
+                                        obt->obt_qctxt.lqc_flags &=
+                                                ~UGQUOTA2LQC(oqctl->qc_type);
+                                else if (quota_is_off(qctxt, oqctl))
+                                                rc = -EALREADY;
+                        }
+                        cfs_up(&obt->obt_quotachecking);
                 }
+
+                break;
+        case Q_SETQUOTA:
+                /* currently, it is only used for nullifying the quota */
+                handle = quota_barrier(&obd->u.obt.obt_qctxt, oqctl, 1);
+
+                push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
+                rc = fsfilt_quotactl(obd, obd->u.obt.obt_sb, oqctl);
+
+                if (!rc) {
+                        oqctl->qc_cmd = Q_SYNC;
+                        fsfilt_quotactl(obd, obd->u.obt.obt_sb, oqctl);
+                        oqctl->qc_cmd = Q_SETQUOTA;
+                }
+                pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
+                quota_unbarrier(handle);
+
+                lqs = quota_search_lqs(LQS_KEY(oqctl->qc_type, oqctl->qc_id),
+                                       qctxt, 0);
+                if (lqs == NULL || IS_ERR(lqs)){
+                        CERROR("fail to create lqs during setquota operation "
+                               "for %sid %u\n", oqctl->qc_type ? "g" : "u",
+                               oqctl->qc_id);
+                } else {
+                        lqs->lqs_flags &= ~QB_SET;
+                        lqs_putref(lqs);
+                }
+
                 break;
         case Q_INITQUOTA:
                 {
-                unsigned int uid = 0, gid = 0;
+                unsigned int id[MAXQUOTAS] = { 0, 0 };
 
                 /* Initialize quota limit to MIN_QLIMIT */
                 LASSERT(oqctl->qc_dqblk.dqb_valid == QIF_BLIMITS);
                 LASSERT(oqctl->qc_dqblk.dqb_bsoftlimit == 0);
 
-                /* There might be a pending dqacq/dqrel (which is going to
-                 * clear stale limits on slave). we should wait for it's 
-                 * completion then initialize limits */
-                qctxt_wait_pending_dqacq(&obd->u.obt.obt_qctxt, 
-                                         oqctl->qc_id, oqctl->qc_type, 1);
-
                 if (!oqctl->qc_dqblk.dqb_bhardlimit)
                         goto adjust;
-                
+
+               /* There might be a pending dqacq/dqrel (which is going to
+                 * clear stale limits on slave). we should wait for it's
+                 * completion then initialize limits */
+                handle = quota_barrier(&obd->u.obt.obt_qctxt, oqctl, 1);
                 LASSERT(oqctl->qc_dqblk.dqb_bhardlimit == MIN_QLIMIT);
                 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
                 rc = fsfilt_quotactl(obd, obd->u.obt.obt_sb, oqctl);
@@ -157,18 +242,36 @@ int filter_quota_ctl(struct obd_export *exp, struct obd_quotactl *oqctl)
                         oqctl->qc_cmd = Q_INITQUOTA;
                 }
                 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
+                quota_unbarrier(handle);
 
                 if (rc)
                         RETURN(rc);
 adjust:
+                lqs = quota_search_lqs(LQS_KEY(oqctl->qc_type, oqctl->qc_id),
+                                       qctxt, 1);
+                if (lqs == NULL || IS_ERR(lqs)){
+                        CERROR("fail to create lqs during setquota operation "
+                               "for %sid %u\n", oqctl->qc_type ? "g" : "u",
+                               oqctl->qc_id);
+                        break;
+                } else {
+                        lqs->lqs_flags |= QB_SET;
+                        lqs_putref(lqs);
+                }
+
                 /* Trigger qunit pre-acquire */
                 if (oqctl->qc_type == USRQUOTA)
-                        uid = oqctl->qc_id;
+                        id[USRQUOTA] = oqctl->qc_id;
                 else
-                        gid = oqctl->qc_id;
+                        id[GRPQUOTA] = oqctl->qc_id;
+
+                rc = qctxt_adjust_qunit(obd, &obd->u.obt.obt_qctxt,
+                                        id, 1, 0, NULL);
+                if (rc == -EDQUOT || rc == -EBUSY) {
+                        CDEBUG(D_QUOTA, "rc: %d.\n", rc);
+                        rc = 0;
+                }
 
-                rc = qctxt_adjust_qunit(obd, &obd->u.obt.obt_qctxt, 
-                                        uid, gid, 1, 0);
                 break;
                 }
         default:
@@ -176,86 +279,132 @@ adjust:
                        obd->obd_name, oqctl->qc_cmd);
                 RETURN(-EFAULT);
         }
+        cfs_gettimeofday(&work_end);
+        timediff = cfs_timeval_sub(&work_end, &work_start, NULL);
+        lprocfs_counter_add(qctxt->lqc_stats, LQUOTA_QUOTA_CTL, timediff);
 
         RETURN(rc);
 }
 #endif /* __KERNEL__ */
+#endif
 
-int client_quota_ctl(struct obd_export *exp, struct obd_quotactl *oqctl)
+int client_quota_ctl(struct obd_device *unused, struct obd_export *exp,
+                     struct obd_quotactl *oqctl)
 {
-        struct ptlrpc_request *req;
-        struct obd_quotactl *oqc;
-        int size[2] = { sizeof(struct ptlrpc_body), sizeof(*oqctl) };
-        int ver, opc, rc;
+        struct ptlrpc_request   *req;
+        struct obd_quotactl     *oqc;
+        const struct req_format *rf;
+        int                      ver, opc, rc;
         ENTRY;
 
         if (!strcmp(exp->exp_obd->obd_type->typ_name, LUSTRE_MDC_NAME)) {
+                rf  = &RQF_MDS_QUOTACTL;
                 ver = LUSTRE_MDS_VERSION,
                 opc = MDS_QUOTACTL;
         } else if (!strcmp(exp->exp_obd->obd_type->typ_name, LUSTRE_OSC_NAME)) {
+                rf  = &RQF_OST_QUOTACTL;
                 ver = LUSTRE_OST_VERSION,
                 opc = OST_QUOTACTL;
         } else {
                 RETURN(-EINVAL);
         }
 
-        req = ptlrpc_prep_req(class_exp2cliimp(exp), ver, opc, 2, size, NULL);
-        if (!req)
-                GOTO(out, rc = -ENOMEM);
+        req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp), rf, ver, opc);
+        if (req == NULL)
+                RETURN(-ENOMEM);
 
-        oqc = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF, sizeof(*oqctl));
+        oqc = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
         *oqc = *oqctl;
 
-        ptlrpc_req_set_repsize(req, 2, size);
+        ptlrpc_request_set_replen(req);
+        ptlrpc_at_set_req_timeout(req);
+        req->rq_no_resend = 1;
 
         rc = ptlrpc_queue_wait(req);
-        if (!rc) {
-                oqc = lustre_swab_repbuf(req, REPLY_REC_OFF, sizeof(*oqc),
-                                         lustre_swab_obd_quotactl);
-                if (oqc == NULL) {
-                        CERROR ("Can't unpack obd_quotactl\n");
-                        GOTO(out, rc = -EPROTO);
-                }
+        if (rc) {
+                CERROR("ptlrpc_queue_wait failed, rc: %d\n", rc);
+                GOTO(out, rc);
+        }
+
+        oqc = NULL;
+        if (req->rq_repmsg)
+                oqc = req_capsule_server_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
 
-                *oqctl = *oqc;
+        if (oqc == NULL) {
+                CERROR ("Can't unpack obd_quotactl\n");
+                GOTO(out, rc = -EPROTO);
         }
+
+        *oqctl = *oqc;
+        EXIT;
 out:
         ptlrpc_req_finished(req);
-        RETURN (rc);
+
+        return rc;
 }
 
-int lov_quota_ctl(struct obd_export *exp, struct obd_quotactl *oqctl)
+/**
+ * For lmv, only need to send request to master MDT, and the master MDT will
+ * process with other slave MDTs.
+ */
+int lmv_quota_ctl(struct obd_device *unused, struct obd_export *exp,
+                  struct obd_quotactl *oqctl)
+{
+        struct obd_device *obd = class_exp2obd(exp);
+        struct lmv_obd *lmv = &obd->u.lmv;
+        struct lmv_tgt_desc *tgt = &lmv->tgts[0];
+        int rc;
+        ENTRY;
+
+        if (!lmv->desc.ld_tgt_count || !tgt->ltd_active) {
+                CERROR("master lmv inactive\n");
+                RETURN(-EIO);
+        }
+
+        rc = obd_quotactl(tgt->ltd_exp, oqctl);
+        RETURN(rc);
+}
+
+int lov_quota_ctl(struct obd_device *unused, struct obd_export *exp,
+                  struct obd_quotactl *oqctl)
 {
         struct obd_device *obd = class_exp2obd(exp);
         struct lov_obd *lov = &obd->u.lov;
+        struct lov_tgt_desc *tgt;
         __u64 curspace = 0;
-        __u32 bhardlimit = 0;
+        __u64 bhardlimit = 0;
         int i, rc = 0;
         ENTRY;
 
-        if (oqctl->qc_cmd != Q_QUOTAON && oqctl->qc_cmd != Q_QUOTAOFF &&
-            oqctl->qc_cmd != Q_GETOQUOTA && oqctl->qc_cmd != Q_INITQUOTA) {
+        if (oqctl->qc_cmd != LUSTRE_Q_QUOTAON &&
+            oqctl->qc_cmd != LUSTRE_Q_QUOTAOFF &&
+            oqctl->qc_cmd != Q_GETOQUOTA &&
+            oqctl->qc_cmd != Q_INITQUOTA &&
+            oqctl->qc_cmd != LUSTRE_Q_SETQUOTA &&
+            oqctl->qc_cmd != Q_FINVALIDATE) {
                 CERROR("bad quota opc %x for lov obd", oqctl->qc_cmd);
                 RETURN(-EFAULT);
         }
 
+        /* for lov tgt */
+        obd_getref(obd);
         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
                 int err;
 
-                if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_active) {
+                tgt = lov->lov_tgts[i];
+                if (!tgt || !tgt->ltd_active || tgt->ltd_reap) {
                         if (oqctl->qc_cmd == Q_GETOQUOTA) {
+                                rc = -EREMOTEIO;
                                 CERROR("ost %d is inactive\n", i);
-                                rc = -EIO;
-                                break;
                         } else {
                                 CDEBUG(D_HA, "ost %d is inactive\n", i);
-                                continue;
                         }
+                        continue;
                 }
 
-                err = obd_quotactl(lov->lov_tgts[i]->ltd_exp, oqctl);
+                err = obd_quotactl(tgt->ltd_exp, oqctl);
                 if (err) {
-                        if (lov->lov_tgts[i]->ltd_active && !rc)
+                        if (tgt->ltd_active && !rc)
                                 rc = err;
                         continue;
                 }
@@ -265,6 +414,7 @@ int lov_quota_ctl(struct obd_export *exp, struct obd_quotactl *oqctl)
                         bhardlimit += oqctl->qc_dqblk.dqb_bhardlimit;
                 }
         }
+        obd_putref(obd);
 
         if (oqctl->qc_cmd == Q_GETOQUOTA) {
                 oqctl->qc_dqblk.dqb_curspace = curspace;
@@ -272,4 +422,3 @@ int lov_quota_ctl(struct obd_export *exp, struct obd_quotactl *oqctl)
         }
         RETURN(rc);
 }
-